Phenotypic similarity based on g2d, g2p _across_ SSPOs about monarch-semantic-similarity-profiles HOT 9 CLOSED

@souzadevinicius can you try to use phenio.db as located in the URL @kevinschaper provided?

from monarch-semantic-similarity-profiles.

To clarify: I didn't intend to suggest not using jaccard, I just pointed out that jaccard is between a pair of phenotype terms is computed using the ontology only, the associations make no difference.

IC can be calculated with different corpuses: just the ontology, ontology + associations.

from monarch-semantic-similarity-profiles.

@souzadevinicius I updated the text for this issue!

from monarch-semantic-similarity-profiles.

cc @kevinschaper

from monarch-semantic-similarity-profiles.

I'm working on subsetted tsv output this week, so soon I'll have an "all gene to phenotype" file or a "just mouse gene to phenotype" file.

I don't know how it would mix in to your pipeline, but the phenio.db that's distributed with monarch-kg builds has all of the phenotype associations populated in the term_association table.

It also might be practical (if ugly) to grep from the tsv:

zcat monarch-kg-denormalized-edges.tsv.gz | grep -e '^id' -e 'has_phenotype' | grep 'MGI:' | grep 'MP:'

or a bit more structured by using the sqlite artifact:

sqlite3 monarch-kg.db -cmd ".mode tabs" -cmd ".headers on"  "select * from denormalized_edges where subject_taxon = 'NCBITaxon:10090' and predicate = 'biolink:has_phenotype'" > mgi_g2p.tsv

from monarch-semantic-similarity-profiles.

@kevinschaper how did the phenio build end up with all of those associations? is this phenio + all of monarch KG? Any documentation (or issue) about this, and a location where to get it?

I also learned something else today from @cmungall. @souzadevinicius we should typically not use Jaccard for comparing phenotypic profiles with gene associations. It seems that we should be using the "Resnik score" only, which somehow (the details are hazy) takes these scores into account by leveraging Information Content. Can you, for our next meeting:

1. Understand how semsimian computes resnik (formula, how it works, with an example)
2. Check if the resnik scores look somehow "sane" when using semsimian?

from monarch-semantic-similarity-profiles.

Sorry, not great documentation. Here is the issue and where it happens in the code: https://github.com/monarch-initiative/monarch-ingest/blob/d65456eb3667a47d960e21f766b1ec65f1b4f774/scripts/load_sqlite.sh#L36

Semsimian uses phenio.db directly, and the semantic sql schema has a table for term associations, so it made sense to supply the associations that way...except that it's obviously a bit weird and circular to populate monarch-kg with phenio and then populate phenio with associations from monarch-kg. So the solution we have right now is that a kg release includes a phenio.db with these extra associations.

from monarch-semantic-similarity-profiles.

Oh, kg artifacts available at: http://data.monarchinitiative.org/monarch-kg/latest/ (purl still to come, of course...)

from monarch-semantic-similarity-profiles.

This issue has been resolved - we now understand that the gene associations were meant to affect only the IC score, and not the Jaccard score. Thanks everyone for your help!

from monarch-semantic-similarity-profiles.